1. Technical Field
The present invention relates to an image processing apparatus, an image processing method, and a computer readable-medium.
2. Related Art
In the field of image processing, a technology has been known which detects a region of interest (ROI) from an image to be processed, by generating a plurality of feature maps representing physical features obtained from the image and integrating them.
As a more concrete method, for example, according to Treisman's feature integration theory, a saliency map obtained by integrating a plurality of feature maps is obtained as a region of interest of an image. According to the theory of Itti, Koch et al., feature maps are normalized by the square error from the average value of each, and all the feature maps are integrated by the linear combination method. Then, the integrated feature map is recursively filtered by a difference-of-Gaussian filter, and a local peak of the finally obtained filtering result is obtained as a region of interest (see e.g., Non-patent Reference 1).
For the plurality of feature maps used by the above-mentioned theories, it is considered typical to use the nature of paying attention with higher priority to a visually salient object in the initial stage of the human viewing process such as high lightness, a specific color or high contrast.
[Non-patent Reference 1] Laurent Itti, Christof Koch, and Ernst Niebur, “A Model of Saliency-Based Visual Attention for Rapid Scene Analysis”, IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, (U.S.A.), IEEE Computer Society, November 1998, Vol. 20, Number 11, pp. 1254-1259
However, in the case of the theory described in Non-patent Reference 1, the weight on the kind of the physical feature is fixed, and when it is applied, for example, to an image having various physical features, it is difficult to appropriately extract a region of interest from the image.